Analyzing the Results

_images/bmtk_architecture_analyzer_highlight.jpg

After a simulation has completed BMTK will automatically save the results to the output folder. It is possible to have BMTK read and analyze the results before the simulation has exited but usually not required. The type of results saved during the simulation is determined by the “reports” section of the simulation config. But most commonly it is a spikes-trains file, a cell-variable report, an extracellular potential recording, and in the case of PopNet a report of the firing rate dynamics.

The output files follow the SONATA Data format, and tools like pySONATA or libSONATA can be used to read them. The following consists of a high-level overview of these different formats and how you can use BMTK to find the results.

Spike-Trains

Contains action-potentials/spikes for all the nodes within a given population. In the HDF5 the spikes are stored under /spikes/_<population_name>_/ and contains two data-sets:

  • timestamps (size N): A list of all the spikes during the simulation

  • node_ids (size N): A list of corresponding node_ids for each spike

_images/spikes_report.png

The group may contain an attribute “sorting” with values none, by_id, and by_time to indicate if and how the spikes are sorted (default none). There is also an optional attribute “units” for the timestamps, but defaults to milliseconds.

Reading

You can use the SpikeTrains class to load a sonata spikes files (or csv or nwb) into memory:

from bmtk.utils.reports.spike_trains import SpikeTrains

pop_name = ...
spikes = SpikeTrains.load('output/spikes.h5', population='pop_name')
assert(spikes.populations = [pop_name])
print(spikes.n_spikes())
print(spikes.node_ids())

If you know what node you want use the get_times() method to return an array of all spikes for a single node

node0_times = spikes.get_times(node_id=0)
print(node0_times)

Otherwise you can use the to_dataframe() or spikes() method to get a list of all node_ids plus timestamps

for ts, pop, node_id in spikes():
   assert(pop == pop_name)
   ...

print(spikes.to_dataframe())

Plotting

Analysis

Creating Spike Trains

Commonly it’s necessary to generate spike-trains to use as inputs for a simulation. One option is to use FilterNet to generate inputs from external stimuli. You can also use the output of one simulation as the input to the next. But if you need to generate your own spike files, in property SONATA format, BMTK provides two ways of readily doing so. One is to use the SpikeTrains class add_spikes() or add_spike() method:

from bmtk.utils.reports.spike_trains import SpikeTrains

spikes = SpikeTrains(population='my_inputs')
spikes.add_spikes(node_ids=0, timestamps=[1.0, 2.0, 3.0, ...])
spikes.add_spike(node_id=[1, 1, 2, 3],
                 timestamps=[0.5, 0.9, 1.0, 1.0])

Or use the PoissonSpikeGenerator class

from bmtk.utils.reports.spike_trains import PoissonSpikeGenerator

psg = PoissonSpikeGenerator(population='thalamus')
times = np.linspace(0.0, 3.0, 100000)
frs = 10*np.sin(times) + 5.0
psg.add(node_ids=range(10), firing_rate=frs, times=times)
psg.add(node_ids=range(10, 20), firing_rate=15.0, times=(0.0, 3.0))
psg.to_sonata('./inputs/thamlamus_inputs.h5')

Cell Variable Report

Used to record the traces of intracellular and membrane variables over the course of the simulation, like membrane potential V. In the HDF5 cell-reports are stored under /report/<population_name>/ with the most relevant datasets:

  • data (size T_times x N_segments): All the recorded values, each row a different step in time and each column a different segment/cell

  • mapping/time (size 3 or T_times): For the exact times of each recording. If the simulation time steps are uniform then the dataset contains 3 values: start_time, stop_time, and time_step (all in ms). Otherwise the will be of size T_times for each recording time since start of simulation.

  • node_ids: used to map each column to a specific cell

_images/cell_reports.png

If recording is done on point-neurons or one is only recording from the soma, there will be one column in “data” for each node. If recording different sections from a multi-compartmental neuron then mapping/index_pointers should be used:

_images/segmentation_indexing.jpg

Reading Cell Variables

The CompartmentReport class should be used to pull data from a cell report.

bmtk.utils.reports.compartment import CompartmentReport

pop_name = ...
report = CompartmentReport('output/membrane_vm.h5',
population=pop_name, mode='r')
print(report.variable())
print(report.tstart(), report.tstop(), report.dt())
print(report.element_ids(node_id=0))
print(report.element_pos(node_id=0))
print(report.data(node_id=0))

v
0.0, 3.0, 0.01
[0, 1, 2, 2, ...]
[0.5, 0.5, 0.5, 05, ...]
[[-50.010, -50.911, -50.995, -51.000, ...]
     [-49.933, -49.330, -50.011, -50.667, ...]
     ...]

Extracellular Potential

Rates